Expanding the Utility of Pyhfo: Lightweight Deep Learning-powered End-to-end High-frequency Oscillations Analysis Application
Abstract number :
3.293
Submission category :
3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year :
2024
Submission ID :
500
Source :
www.aesnet.org
Presentation date :
12/9/2024 12:00:00 AM
Published date :
Authors :
Presenting Author: Yuanyi Ding, MS – University of California, Los Angeles
Yipeng Zhang, Dr. – UCLA
Lawrence Liu, BS – University of California, Los Angeles
Tonmoy Monsoor, PhD – University of California, Los Angeles
Atsuro Daida, MD,PhD – UCLA Mattel Children's Hospital
Yun Zhang, BS – University of California, Los Angeles
Yuqi Huang, BS – University of California, Los Angeles
Shingo Oana, MD, PhD – University of California, Los Angeles
Sotaro Kanai, MD, PhD – Tottori University, Faculty of Medicine
Shaun Hussain, MD, MS – UCLA Mattel Children's Hospital, David Geffen School of Medicine
Raman Sankar, MD, PhD – University of California, Los Angeles
Aria Fallah, MD, MS – UCLA Mattel Children's Hospital
Jerome Engel Jr., MD, PhD – University of California, Los Angeles
Richard Staba, PhD – University of California, Los Angeles
William Speier, PhD – University of California, Los Angeles,
Vwani Roychowdhury, PhD – UCLA
Hiroki Nariai, MD, PhD, MS – UCLA Mattel Children's Hospital
Rationale: This study develops and validates PyHFO, an end-to-end software platform that applies deep learning techniques to detect/classify high-frequency oscillations (HFOs) from EEG recordings.
Methods: PyHFO features a multi-window Graphical User Interface (GUI) for efficient HFO analysis. It presents a user-friendly and intuitive interface that caters to both technical and non-technical users, streamlining the process of HFO detection, classification, and user annotation (Figure 1). PyHFO operates through four primary stages: EEG signal reading, data filtering, HFO detection, and Deep Learning (DL)-based HFO classification. The output of this pipeline includes detected events based on the short-term energy (STE) and Montreal Neurological Institute and Hospital (MNI) detectors, accompanied by annotations of real HFOs, artifacts, and HFOs with spikes, generated by pre-trained DL-based HFO classification models. pyHFO also allows users to modify or refine these annotations based on their assessment.
Results: Validation of PyHFO was conducted using three datasets: one with grid/strip electrodes (UCLA iEEG Dataset), another combining grid/strip and depth electrodes (Zurich iEEG), and a third from rodent studies involving depth electrodes in the neocortex and hippocampus (UCLA Rodent Dataset). PyHFO demonstrated an ability to handle datasets efficiently, with optimization techniques enabling it to achieve speeds up to 50 times faster than traditional HFO detection applications and offering high implementation accuracy relative to most mainstream publicly released software (Table 1). Users have the flexibility to employ our pre-trained DL model or train custom models with their own EEG data. The latest version of the pyHFO software can be downloaded on the website below:
https://github.com/roychowdhuryresearch/pyHFO/releases
Conclusions: PyHFO successfully bridges the computational challenge faced in applying DL techniques to EEG data analysis in epilepsy studies, presenting a feasible solution for both clinical and research settings. By offering a user-friendly and computationally efficient platform, PyHFO paves the way for the broader adoption of advanced EEG data analysis tools in clinical practice and fosters the potential for large-scale research collaborations. Additionally, we plan to expand the existing pipeline to include capabilities for spike, spindle, intra-operative ECoG, and scalp EEG analysis in the near future.
Funding: The National Institute of Health, K23NS128318
Neurophysiology